Rigid3D typically stores these as a unit quaternion (for rotation) and a 3-vector (for translation). For this tutorial, we'll use the Sophus::SE3d (C++) or scipy.spatial.transform.Rotation (Python). If you're working with ROS, tf2::Transform is analogous. C++ (Eigen + Sophus) #include <sophus/se3.hpp> #include <Eigen/Core> #include <iostream> using Sophus::SE3d; using Eigen::Vector3d; using Eigen::Quaterniond; Python (NumPy + SciPy or transforms3d) import numpy as np from scipy.spatial.transform import Rotation as R # Or use `from transforms3d.quaternions import quat2mat` 3. Creating a Rigid3D Object Let’s create a transformation that represents: rotate 90° about Z-axis, then translate by (1, 0, 0).
[ T_ac = T_ab \cdot T_bc ]
# Rotation: 90 deg around Z r = R.from_euler('z', 90, degrees=True) t = np.array([1.0, 0.0, 0.0]) T = np.eye(4) T[:3,:3] = r.as_matrix() T[:3, 3] = t 4. Applying the Transformation Transform a 3D point ( p = (0, 1, 0) ) from frame A to frame B. rigid3d tutorial
T_ac = T_ab @ T_bc Order matters. The rightmost transform is applied first to the point. 6. Inversion The inverse of a rigid transform ( T_ab ) is ( T_ba ). It rotates by ( R^T ) and translates by ( -R^T t ).
In robotics, computer vision, and 3D graphics, the ability to represent rotations and translations in 3D space is fundamental. The Rigid3D object (often found in libraries like Sophus , Eigen , geometry_msgs , or tf2 ) is the industry-standard way to do this. Unlike a 4x4 homogeneous matrix, Rigid3D separates rotation (SO(3)) and translation, offering better numerical stability and mathematical clarity. Rigid3D typically stores these as a unit quaternion
T_ba = np.linalg.inv(T_ab) # For rigid transforms, this is more efficient: R_inv = T_ab[:3,:3].T t_inv = -R_inv @ T_ab[:3,3] C++:
[ p_B = R_AB \cdot p_A + t_AB ]
SE3d T_ba = T_ab.inverse();